Skip to content

title: CLI Reference description: Complete reference for the ciagen CLI : generate, evaluate, filter, and caption commands. keywords: cli, command line, reference

CLI Reference

The ciagen command-line tool provides access to all core operations.

Global

ciagen --help

ciagen generate

Generate synthetic images from real images.

ciagen generate \
    --source DIR \
    --output DIR \
    --extractor NAME \
    --sd-model MODEL \
    --cn-model MODEL \
    --num N \
    --seed SEED \
    --device DEVICE \
    --prompt TEXT \
    --negative-prompt TEXT \
    --quality N \
    --guidance-scale FLOAT
Flag Required Default Description
--source Yes : Source images directory
--output Yes : Output directory
--extractor Yes : canny, openpose, segmentation, mediapipe_face
--sd-model Yes : Stable Diffusion model ID
--cn-model Yes : ControlNet model ID
--num No 1 Images per source image
--seed No 34567 Random seed
--device No cuda cuda or cpu
--prompt No : Positive prompt
--negative-prompt No : Negative prompt
--quality No 30 Inference steps
--guidance-scale No 7.0 Classifier-free guidance scale

ciagen evaluate

Compute quality metrics for generated images.

ciagen evaluate \
    --real DIR \
    --generated DIR \
    --metrics fid mld \
    --feature-extractor vit \
    --batch-size 32 \
    --device cuda
Flag Required Default Description
--real Yes : Real images directory
--generated Yes : Generated images directory
--metrics No fid mld Metrics to compute
--feature-extractor No vit vit or inception
--batch-size No 32 Batch size
--device No auto cuda or cpu

ciagen filter

Filter generated images by quality score.

ciagen filter \
    --generated DIR \
    --method METHOD \
    --value VALUE \
    --metric mld \
    --feature-extractor vit
Flag Required Default Description
--generated Yes : Generated images directory
--method Yes : threshold, top-k, top-p
--value Yes : Threshold value
--metric No mld PTD metric name
--feature-extractor No vit Feature extractor name

ciagen caption

Generate captions for images using a vision model.

ciagen caption \
    --images DIR \
    --output DIR \
    --engine ENGINE \
    --model MODEL \
    --api-key KEY
Flag Required Default Description
--images Yes : Images to caption
--output Yes : Output captions directory
--engine No openai openai or ollama
--model No gpt-4o-mini Vision model name
--api-key No : API key (for OpenAI)